國立虎尾科技大學 |

Theory and Algorithms for Dynamic and Adaptive Online Learning.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Theory and Algorithms for Dynamic and Adaptive Online Learning./
作者:	Yang, Scott.
面頁冊數:	1 online resource (221 pages)
附註:	Source: Dissertation Abstracts International, Volume: 78-12(E), Section: B.
Contained By:	Dissertation Abstracts International78-12B(E).
標題:	Applied mathematics. -
電子資源:	click for full text (PQDT)
ISBN:	9780355128666

Theory and Algorithms for Dynamic and Adaptive Online Learning.
Yang, Scott.

Theory and Algorithms for Dynamic and Adaptive Online Learning. - 1 online resource (221 pages)

Source: Dissertation Abstracts International, Volume: 78-12(E), Section: B.

Thesis (Ph.D.)--New York University, 2017.

Includes bibliographical references

Online learning is a powerful and flexible model for sequential prediction. In this model, algorithms process one sample at a time with an update per iteration that is often computationally cheap and simple to implement. As a result, online learning algorithms have become an attractive solution for modern machine learning applications with very large data sets.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2018

Mode of access: World Wide Web

ISBN: 9780355128666Subjects--Topical Terms:

1069907
Applied mathematics.
Index Terms--Genre/Form:

554714
Electronic books.

Theory and Algorithms for Dynamic and Adaptive Online Learning.
LDR:03854ntm a2200397Ki 4500 001 916763
005 20180928111500.5
006 m o u
007 cr mn||||a|a||
008 190606s2017 xx obm 000 0 eng d
020 $a 9780355128666
035 $a (MiAaPQ)AAI10261692
035 $a (MiAaPQ)nyu:12961
035 $a AAI10261692
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Yang, Scott. $3 1190587
245 1 0 $a Theory and Algorithms for Dynamic and Adaptive Online Learning.
264 0 $c 2017
300 $a 1 online resource (221 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertation Abstracts International, Volume: 78-12(E), Section: B.
500 $a Adviser: Mehryar Mohri.
502 $a Thesis (Ph.D.)--New York University, 2017.
504 $a Includes bibliographical references
520 $a Online learning is a powerful and flexible model for sequential prediction. In this model, algorithms process one sample at a time with an update per iteration that is often computationally cheap and simple to implement. As a result, online learning algorithms have become an attractive solution for modern machine learning applications with very large data sets.
520 $a The standard benchmark adopted in online learning is external regret, which is the difference between the cumulative loss of an algorithm and that of the best static model in hindsight. Remarkably, there exist online learning algorithms for which the average external regret converges to zero, even when the learner receives only limited information about the environment or the data-generating distribution. However, this measure of performance may not be useful when the data is non-stationary and no static model admits a small loss. It may also be inadequate when a learner's algorithm is not robust and a simple perturbation of the algorithm's predictions could have been substantially more effective. By construction, online learning algorithms that are designed to minimize external regret cannot be expected to perform well in dynamic environments.
520 $a Most online learning algorithms are also designed with worst-case guarantees in mind, so that they will guarantee a minimum level of performance. This often makes their updates and predictions overly conservative, so that they are unable to adapt to and learn faster from easier data. Finally, while model selection has been well-developed in the offline setting using ideas such as structural risk minimization, its treatment is lacking in the online learning setting.
520 $a This dissertation addresses all the issues just mentioned and presents online learning algorithms that achieve more compelling guarantees by adjusting for dynamic environments and adapting to the given data. In particular, it develops: a novel and flexible framework that can be used to design efficient online algorithms with strong guarantees against non-stationary and evolving benchmarks; a new algorithm that can guarantee the robustness of the learners' predictions and that is motivated by insightful game theoretic considerations; a new and general framework for online convex optimization that can leverage a learner's predictions and tune an algorithm's performance to adaptively optimize for those predictions; a new theoretical analysis and algorithmic design for model selection applied to online learning that is consistent with the offline scenario.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2018
538 $a Mode of access: World Wide Web
650 4 $a Applied mathematics. $3 1069907
650 4 $a Computer science. $3 573171
650 4 $a Statistics. $3 556824
650 4 $a Artificial intelligence. $3 559380
655 7 $a Electronic books. $2 local $3 554714
690 $a 0364
690 $a 0984
690 $a 0463
690 $a 0800
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a New York University. $b Mathematics. $3 1180487
773 0 $t Dissertation Abstracts International $g 78-12B(E).
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10261692 $z click for full text (PQDT)